Discussion Paper

No.1796

September 2021

Communication within firms: evidence from CEO turnovers

Stephen Michael Impink Andrea PratRaffaella Sadun

ISSN 2042-2695

Abstract This paper uses novel, firm-level measures derived from communications metadata before and after a CEO transition in 102 firms to study if CEO turnover impacts employees’ communication flows. We find that CEO turnover leads to an initial decrease in intra-firm communication, followed by a significant increase approximately five months after the CEO change. The increase is driven primarily by vertical (i.e. manager to employee) communication. Greater increases in communication after CEO change are associated with greater increases in firm market returns.

Key words: CEO change, communication, alignment

We thank Wouter Dessein, Stephen Hansen, Mitch Hoffman, Adam Kleinbaum, Niko Matouschek, Meg Meyer, Antoinette Schoar, Iwan Barankay, Robert Gibbons, Toby Stuart, seminar participants at Columbia, and participants at the Spring 2021 NBER Organizational Economics, 2021 Strategy Science (HBS), 2021 Ghoshal (LBS) Conferences for insightful feedback.

Stephen Michael Impink, New York University. Andrea Prat, Columbia University. Raffaella Sadun, Harvard University and Centre for Economic Performance, London School of Economics.

This paper was produced as part of the Centre’s Growth Programme. The Centre for Economic Performance is financed by the Economic and Social Research Council.

Published by Centre for Economic Performance London School of Economics and Political Science Houghton Street London WC2A 2AE

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means without the prior permission in writing of the publisher nor be issued to the public or circulated in any form other than that in which it is published.

Requests for permission to reproduce any article or part of the Working Paper should be sent to the editor at the above address.

S.M. Impink, A. Prat and R. Sadun, submitted 2021.

2

1. Introduction

There is considerable differentiation in what CEOs do, and ultimately, the extent to which they are able to

affect organizational performance (Bertrand & Schoar 2003, Bandiera et al. 2020). However, the specific

mechanisms through which CEO decision making and behavior affect organizations are still largely

unknown. In this paper, we study a specific channel through which CEOs may be able to affect firm

performance: that is, by shaping internal communication flows.

A large theoretical literature has emphasized the importance of communication in firms. For

instance, intra-firm communication is essential to developing and executing firms’ strategies, enabling

firms to share knowledge, create organizational memory, and make decisions (Simon 1947, Arrow 1974).

More efficient and effective communication can reduce transaction costs in decisions to internalize aspects

of operations (Williamson 1979), overcome the cognitive limitations of employees (Simon 1947), increase

the ability for firms to analyze and understand information (Arrow 1974), make “sense” out of changes in

firm strategy (Weick 1995), and is one of the main reasons why firms form (Coase 1937). In multi-

divisional firms, employees use communication to coordinate decision-making across departments and

align those decisions to local conditions to achieve the firm’s goals (Alonso et al. 2008, Marshak & Radner

1972).

The literature has also long recognized that communication patterns across firms are not random,

but can be shaped by leaders. For example, if intra-firm communication is crucial to the firm strategy, we

should expect corporate leaders to attempt to shape it. This fact is recognized by management scholars (e.g.

Kotter 1995, Schein 1994) and economists (e.g., Rotemberg and Saloner 2000, Van den Steen 2005). In

spite of this rich theoretical literature, however, the direct connection between leaders and communication

patterns has not yet been studied empirically due to data limitations. Comparable observational data on

firm-level communication for large samples of firms are hard to find, and longitudinal data following the

evolution of communication patterns over time within firms has not, to our knowledge, been available to

researchers (Impink, Prat & Sadun 2020).

3

This paper provides the first empirical study of how CEOs affect internal communication patterns

across a large sample of firms leveraging unique meeting and email metadata, from 102 firms undergoing

a leadership change. CEO turnover is a common organizational “shock” impacting a firm’s strategy and

performance. We examine whether and how CEO turnover impacts employee’s formal, internal

communication patterns (as measured by intra-firm emails and meetings) with an event study research

design that examines communication flows from six months before the transition to 14 months after the

transition.

We find that CEO change leads to an initial decrease in intra-firm communication, followed by a

significant increase in intra-firm communication approximately four months after the transition. Second,

the medium-term increase in communication is primarily vertical—i.e. involving communication between

managers to employees and vice-versa—rather than horizontal—i.e. involving managers with managers

and employees with employees. Finally, though we cannot observe the CEO type directly, and we have

very limited performance data, we observe that firms that experience a greater increase in medium-term

communication also have higher cumulative abnormal stock returns for six months after the transition.

We then interpret these findings through the lens of a theoretical model of intra-firm

communication based on Alonso et al. (2008), which formalizes the idea that directly following a CEO

change confusion around the firm’s changing goals and management’s expectations increases ambiguity

around the firm’s objectives, reducing the formal communication needed to coordinate decision-making.

However, as the new CEO sets expectations, redefines priorities, and then shares this new strategy with the

firm, uncertainty is reduced. As uncertainty diminishes, employees within the firm communicate more to

coordinate decisions. CEOs who are better leaders can restore coordination more quickly (Kotter 1995,

Schein 1994) and to a higher point than previously attained.

This study innovates on the prior literature on four fronts. First, our findings contribute to the

literature dedicated to understanding the impact that CEOs have on the “softer,” less quantifiable aspects

of firms that are typically hard to observe (Gibbons & Henderson 2012, Bloom, Sadun, et al. 2012, 2016,

4

Impink, Prat and Sadun, 2020) and their contribution to performance variation across firms (Syverson

2011). Second, we are the first study to use longitudinal firm-level communications metadata to analyze

the impact of an organizational change across many firms. Our research design allows us to examine the

dynamics of intra-firm communication patterns before and after a CEO transition. To our knowledge, only

one other paper, Srivastava et al. (2015), shows the impact of an organizational event on communication

patterns, yet only focuses on a single firm.

Third, we are among the first studies to use communications metadata at scale across many firms

(Polzer et al. 2016, Polzer & DeFilippis 2020, DeFilippis, Impink, Singell, Polzer & Sadun 2020, Jacobs &

Watt 2021). Communications metadata have been used in prior research to analyze employee coordination

(Kleinbaum et al. 2008, Kleinbaum & Stuart 2013), the implementation of corporate strategies (Kleinbaum

& Stuart 2014), and employee-level productivity (Aral et al. 2012). These studies have pioneered the idea

that email and meeting metadata contain valuable insight into information flows within firms and that

changes to information flows have important decision-making implications. However, they usually only

observe a single organization rather than a large cross-section of firms.

Lastly, this paper relates to a rich literature in organizational economics that focuses on the importance

of intra-firm communication (Hart and Holmstrom 2002, Dessein & Santos 2006, Cremer et al. 2007,

Alonso et al. 2008, Rantakari 2008, Calvó‐Armengol et al. 2015, Dessein et al. 2016, Van den Steen 2017)

as a proxy for information flows and knowledge-sharing needed to support effective coordination. This

literature has identified the existence of tradeoffs between the benefits of increased coordination and

employees acting independently. Our contribution is to leverage an established model of intra-firm

communication and decision making (Alonso et al. 2008, Rantakari, 2008) to guide the empirical analysis

of communication flows after a significant organizational event.

The paper proceeds as follows. Section 2 describes the communications metadata and the measures we

developed to test the model’s predictions. Section 3 details the event study and performance research

designs. Section 4 describes the main results and additional findings correlating communication changes

5

with firm performance. Section 5 describes the related theory that we use to understand our findings and

discusses an example of a stylized model used to understand information flows between parties in a multi-

divisional firm. Section 6 concludes.

2. Data

In this section, we describe the data employed in the empirical analysis. We start by describing how we

constructed the sample of firms experiencing a CEO transition event (2.A). We then describe the

communication variables proxying for internal communication flows (2.B) and French-Fama cumulative

returns as stock market performance measures (2.C). Lastly, we discuss some limitations of using

communications metadata (2.D).

2.A. Identifying CEO turnover events

We obtained metadata on meeting and communication flows thanks to a partnership with a large email

provider. The provider allowed us to analyze firm-level aggregates based on meeting and email metadata,

under the condition that the firms used in the sample could not be identified (by other external researchers

or us). To identify a suitable sample for the analysis, the provider gave us a sense of the time period and

geographies for which email and meeting metadata had been captured and retained (approximately three

years of data). We identified all firms experiencing a CEO transition within a three-month window within

this time period using the information on CEO names from Execucomp, Boardex, and Orbis. This sample

consisted of 338 firms, for which we collected additional firm-level data from multiple databases, including

information on the number of employees, industry, CHQ location from Dun & Bradstreet, and revenue data

from Orbis. We manually coded why the CEO transition occurred (e.g., fired for performance issues,

retirement, death, an internal transition to another role, merger, hired by another firm, or left to start an

entrepreneurial venture) from press releases. This dataset was sent to the email provider to be anonymously

6

matched to firms in its database. We describe additional details on the initial sample and matching process

with our email provider in the Appendix in Note A.1.

This matching process led to a sample of 102 firms with communication data before and after a

CEO turnover event. The sample includes firms located in 21 different countries, though 40% of the firms

are located in the United States. The CEOs left for various reasons, including being fired for performance

reasons (20%), retirement (17%), and transfer to another firm (8%). Many CEOs (30%) remained at the

firm in a different position. The majority of the firms in the sample are in the services1 (30%),

manufacturing (19%), or trade (9%) industries. Over half the firms (51%) replaced their CEO with an

internal manager already at the firm. The average firm has 6,545 (SD 4,217) users, with the largest firm

having almost 15,000 users. In any given period, the sample includes roughly 100,000 users. In total, this

communications metadata contains approximately 500M emails and 80M meetings. We provide more

details on the firms in our sample in Table A.6..

2.B. Describing Communication Measures

Once the usable sample was identified, we asked the email provider to build month-level, firm-specific

aggregated communication measures using email and meeting metadata measured before and after the CEO

transition. Our primary analysis is based on variables measuring meetings and email data at the firm-month

level. We were also able to obtain information on emails and meetings for different subsets of each firm,

such as employees by hierarchical level (top managers, i.e. managers of managers; other managers; and

individual contributors) based on each firm’s organogram.2 In addition to total meeting and email

communications, we also received measures of meeting and email communications within and across

functional departments (though only for 89 firms) and measures of meeting communications across similar

1 Includes financial and insurance services. 2 Firms’ organograms are derived from the firms’ listing of formal reporting relationships, which are self-recorded by the firm when the email provider onboards a firm’s employees to the platform. Through this pairing, we can distinguish between employees that are individual contributors, managers, and managers of managers.

7

managerial hierarchy levels (i.e. horizontal communication) and upward with management (i.e. vertical

communication) (91 firms). In the main analysis, we use firm-level aggregates for 102 firms.

Meetings. We received data on average number of meetings per employee, average duration of meetings,

average number of attendees per meeting for 17 months. Table A.7 shows basic summary statistics for the

meeting data. Employees attended, on average, 39 (Median 29, Standard Deviation 32) internal meetings

per month. These meetings lasted about 116 minutes (Med. 79, SD 327) and included 42 attendees (Med

23, SD 62).3 In total, employees were scheduled to attend around 80 hours a month in meetings. We provide

additional information on how these data are aggregated in the Appendix in Note A.2.

We also show the aggregated meeting data along the firm hierarchy, across similar hierarchal levels

(i.e., horizontal communication between employees or between managers) and across different hierarchical

levels between ICs and managers (i.e., vertical communication between managers and workers) to proxy

for the different communication modes discussed in the model. Meetings vary along the hierarchy: senior

managers have more meetings and send more emails than lower-ranked managers and individual

contributors (59 meetings/month (Med. 52, SD 37) for senior managers, 42 meetings/month (Med. 36, SD

28) for other managers, and 20 meetings/month (Med. 16, SD 15) for ICs. Even though managers attend

more meetings, these meetings are on average shorter (Senior managers: 90 minutes/meeting, Med. 77 SD

76; manager: 101 minutes/meeting, Med. 77 SD 145) or than those attended by ICs (151 minutes/meeting,

Med. 85 SD 513). We use t-tests to confirm that the differences between the means of these subgroups are

different from zero. For these groups, the only measure in which the means are not significantly different

is the number of attendees present at a meeting. We report these summary statistics and results in Table A.7

and A.8 under Meetings.

For vertical and horizontal communication flows, there are an average of 14,155 (SD 38,127)

horizontal meetings and 8,722 (SD 29,357) vertical meetings per month at the firm-level (96 firms). This

3 To avoid the inclusion of all-hands meetings and training events, the data provider excluded meetings that lasted longer than eight hours, all-day meetings and multi-day meetings. The data also exclude meetings that were declined by an employee in their calendar metadata.

8

roughly equates to 2 (SD 2) vertical and 5 (SD 4) horizontal meetings per user per firm per month.4 We

report these details in Appendix A in Table A.9 and summarize the data available and provided in Table

A.5.

Emails. For email measures, we received aggregated average number of internal emails, number of

recipients, and the percent of same department only emails for 22 months. We further broke down these

aggregates into the average number of different department emails and the mix of different-same

department emails within each firm. On average, employees sent around 247 (Med. 223, SD 199) internal

emails per month. Senior managers send 333 emails/month (Med. 310, SD 164), other managers: 266

emails/month (Med. 243, SD 130), ICs: 163 emails/month (Med. 132, SD 236). We report these summary

statistics in Table A.7 and Table A.8 under Emails.

2.C. Performance Measures

We measure performance before and after the CEO transition using monthly stock market measures for the

subset of public firms included in our sample (41). Specifically, we build Cumulative Abnormal Returns

(CAR) measures 6 months before and 6 months after the CEO change.5 Table A.10 shows basic summary

statistics of these measures before and after CEO turnover for two alternative CAR measures (Market-

Adjusted CAR, and French-Fama CAR). We provide additional details on the creation of these performance

measures in Appendix A in Note A.3.

4 This measure does not map back to our other data set because these are distinct meetings. For example, even if 100 employees attended an all-hands meeting, then it would still only count as one meeting. 5 The market adjusted CAR is calculated as (return – risk-free market rate). The cumulative Fama-French abnormal stock return is calculated using monthly French-Fama data downloaded from the following website: http://mba.tuck.dartmouth.edu /pages/faculty/ken.french/. We build the pre-turnover CAR measure by summing AR for every month prior to the CEO change (so the CAR for month -6 is the AR between month -7 and -6, for month -5 it is the cumulative sum of the CAR measured at -6, and the AR between month -6 and -5, and so forth. The CAR for the post CEO change time period starts summing AR between month 0 (i.e. the month of the CEO change) and month 1 after the CEO change and so forth. We exclude the CAR at month 0 from our analysis since it reflects the effects of both prior and current CEOs.

9

2.D. Limitation of Communications Metadata.

There are numerous limitations in the metadata aggregates used in the analysis. First, these measures do

not capture all interactions, including informal communications on mobile phones or other consumer-

focused software platforms. Our data only capture formal communications on the email provider’s platform

and do not capture any forms of informal communications (e.g., instant messenger, personal email,

consumer messaging applications, SMS, and many more). Furthermore, employees that deal with highly-

sensitive information may opt for an ad hoc phone call or in-person meeting to avoid sharing the information

in a legally discoverable manner. To the extent that a CEO change affects these margins of adjustment in

communication, we will be unable to capture them in our analysis.

Next, it is unclear if meetings represent what employees actually do or simply what they write in

their digital calendars. For instance, employees could use their calendars inconsistently or may not attend

meetings that they accept in their calendars. We do not know if employees attend the meeting; we only

know that they have not “declined” the meeting. The types of communications, both meetings and email,

could vary drastically across people according to their responding habits. Similarly, for emails, we do not

know if an individual exerted more or less effort writing an email or quickly sending a message. Our data

are limited in that we do not have access to the content or subject lines of emails and meeting requests,

which could provide us with more insight into the interaction and the exertion of effort. For example, a one-

on-one meeting with your manager is very different from a team-building event or monthly all-hands. An

email used to update management is very different from an email used to coordinate firm strategy. Also,

meetings could be set up to work on a single task or many different tasks.

Lastly, we do not have any network measures. Instead, we use firm-level aggregates with some

network connotations (i.e. same versus different departments, horizontal versus vertical communication,

management hierarchy, etc.). Given the nature of these data, the network approach where every person is a

“node” would not adequately anonymize in our sample, and a panel containing more than 0.5 billion

10

communications for the 100K monthly user in our sample would have been difficult and exceedingly costly

to analyze.6

3. Research Design

3. A. Communication

We exploit the CEO transition’s timing by using an event study specification to analyze the change in email

and meeting measures over different months before and after the CEO transition. All data are aggregated

at the firm level and bi-monthly level.7 The CEO transition occurs in month 0, and combined with the first

month after the transition, is denoted “Period 0” in the tables and figures. The base period for these

regressions is the bi-monthly period before the CEO transition (i.e. month -1 and -2, denoted as “Period-1”

in the tables and figures). We include firm-level fixed effects in all specifications, such that the coefficients

represent changes in communication patterns within the same firm over time.

𝑙𝑙𝑙𝑙 (𝑦𝑦𝑖𝑖,𝑡𝑡) = 𝛽𝛽0 + ∑ 𝛽𝛽𝑡𝑡𝐿𝐿−2𝑡𝑡=−3 𝐷𝐷𝑡𝑡 + ∑ 𝛽𝛽𝑡𝑡𝐿𝐿6

𝑡𝑡=0 𝐷𝐷𝑡𝑡 + 𝛾𝛾𝑖𝑖 + ε𝑖𝑖,𝑡𝑡 (1)

Where 𝑦𝑦𝑖𝑖,𝑡𝑡 denotes a communication variable for firm i in bi-month period t (relative to the CEO transition

period), and 𝛾𝛾𝑖𝑖 are firm fixed effects. We cluster standard errors at the firm level. In this specification,

Periods 0 and 1 (i.e. months 0-3 after the CEO transition) map into the short-term transition period, Periods

2-3 (i.e. months 4-7 after the CEO transition) map into medium-term transition period, and Periods 4-6 (i.e.

months 8-12 after the CEO transition) correspond to the new steady state.

Given that CEO changes are likely to be endogenous,8 variation in communication patterns

happening around the time of a CEO transition may be driven by other unobservable and time-varying firm-

level factors (since we control for time-invariant factors with firm-level fixed effects). By observing

6 The information underlying the meeting and email variables contains about 15 petabytes of data. 7 We aggregate the communication data at the bi-monthly data to smooth out measurement error, but the results are qualitatively and quantitatively similar when we use monthly data. See Appendix in Table B.1 (Monthly Data) for more details. 8 In our sample only one CEO transition was driven by truly exogenous events, i.e. CEO illness or death.

11

communications for several months prior to the CEO transition, however, we can at least examine whether

the changes in communication are driven by trends in the data pre-dating the CEO turnover.

3.B. Performance

To examine whether different communication patterns map into different performance outcomes, we

estimate a difference in differences model of the form:

𝑅𝑅𝑖𝑖,𝑡𝑡 = 𝛽𝛽0 + ∑ 𝛽𝛽𝑡𝑡𝐿𝐿3𝑡𝑡=1 𝐷𝐷𝑡𝑡 + ∑ 𝛽𝛽𝑡𝑡𝐿𝐿3

𝑡𝑡=1 𝐷𝐷𝑡𝑡𝑦𝑦�𝑖𝑖 + 𝑦𝑦�𝑖𝑖 + 𝛾𝛾𝑖𝑖 + ε𝑖𝑖,𝑡𝑡 (2)

Where 𝑅𝑅𝑖𝑖,𝑡𝑡 denotes CAR computed for every bi-month following the CEO turnover, starting from Period 0

(corresponding to the CAR month 1, since the CAR in month 0 is excluded from the analysis). As in earlier

specifications, 𝛾𝛾𝑖𝑖 denotes firm level fixed effects, and standard errors are clustered at the firm level. The

variable 𝑦𝑦�𝑖𝑖 is a measure of communication intensity in the medium term after the CEO turnover event. We

focus on medium-term changes in light of Prediction 3, which sees this intermediate time period as the

period in which CEOs’ heterogeneous ability to achieve alignment within the firm is manifested.

4. Results

Main Results. We show the event study results in Exhibit 1 and graph the bimonthly coefficients in the top

two figures in Exhibit 4. Email and meeting measures do not display evidence of a pre-trend in months

before the base period, Period -1, two months before the CEO transition month. Meeting counts and emails

experience a sharp drop immediately after the CEO transition (Period 0, i.e. months 0 and 1 after the

transition), and recover slightly but continue to remain depressed in Period 1, i.e. 2 and 3 months after the

transition (meetings: P0: -33% SD 0.11, P1: -29% SD 0.09; and emails: P0: -16% SD 0.07, P1: -13% SD

0.06). Meetings and emails increase significantly from the base period (Period -1) to Period 2 (i.e. 4 and 5

months after the month of the CEO change) and stabilize through Periods 3 to 6, i.e. between 6 and 12

months after the CEO change month (meeting stabilize ~+20% from base and emails stabilize at ~+10%).

Generally, meetings are more responsive after the CEO change event; Meetings initially decline twice as

12

fast and recover more quickly (recovering in Period 2) compared with emails (recovering in Period 3). The

duration of meetings and the number of attendees also increases in the medium term,

Beyond average effects, we note that the medium-term increase in meetings and emails is more

heterogeneous across firms than the short-term decrease in communication. To illustrate this point, Figure

5.A in Exhibit 5 shows the histograms of two firm level variables: first a measure of the average monthly

change in emails between month 0 and month 2 after the CEO transition, and second, a measure of the

average monthly change in emails between month 2 and month 6 after the transition. The mass of

distribution of medium-term changes is shifted to the right relative to short-term changes depicting a higher

medium-term mean change on average, and the variance of the medium-term changes is larger. We discuss

a possible interpretation of these differences later in this section.

Vertical and Horizontal Communication Interactions. The email provider calculated vertical and horizontal

communication metadata for meetings in 88 firms from six months before until eight months after the CEO

transition.9 We show these results in Exhibit 2 and graph these coefficients in the middle and bottom figures

in Exhibit 4. Similar to the main results, both vertical and horizontal meeting interactions increase starting

in Period 2, four months after the CEO change. We analyze the change in the ratio of vertical to horizontal

meetings over time (i.e. horizontal meeting count divided by vertical meeting counts) to determine if the

mix of intra-firm meeting interaction types changed. The ratio significantly increases in the medium term

(P2 +12% SD 0.02, P3 +18% SD 0.02), i.e. communication flows were more likely to happen across

different hierarchical layers rather than within the same layer.

Robustness Checks. We performed several robustness checks on the results discussed so far, which we

report in Appendix B in Table B.4. First, our data capture all formal communications through an employee’s

work-provided email address. However, some employees may use their work email and calendars for

personal interactions with their colleagues. As such, this additional metadata would be included in the

9 Due to large gaps in the email meta-data from GDPR compliance, the email provider built these measures from meeting meta-data only. Additionally, we are provided with user counts in the second dataset that we used to build the firm x month per user communications measures form the firm.

13

aggregated data we received. Under the assumption that these types of informal interactions are more likely

to be included among employees working in the same department, we built measures of aggregate emails

and meetings based exclusively on data capturing communication flows between employees of different

functional departments.10 Reassuringly, the results are virtually identical when we use this subsample of

interactions (column 3).11

A first robustness check relates to the unbalanced nature of the sample used in the analysis. This is

because some firms only have meeting metadata after the CEO transition (data availability depends on

when firms opted into the email provider’s services). All firms have balanced email data; however, only 55

firms have balanced meeting data, including the transition month. When we re-run the main event study for

the balanced sample of firms with data before and after the transition, we obtain similar coefficients, though

less precisely estimated (column 5).

Second, we investigated whether the changes in communication flows were driven by the timing

of the actual CEO transition or by its mere announcement. We were able to examine the reaction to the

announcement of the CEO transition separately from that of the actual transition for 30 firms, for which the

announcement occurred earlier than a month before the actual CEO transition. We find that the coefficients

on the dummies denoting months after the actual transition are negative and significant (columns 7 and 9)

even in this small subsample of firms, while the point estimates on the dummies relative to the

announcements — while negative — are not precisely estimated (columns 8 and 10).12

10 Kleinbaum et al. (2013), Zhang et al. (2020), for example, show that communication across departments included a higher proportion of communication with weak network ties. This data was only available for 89 firms of the 102 firms in the sample. 11 Kleinbaum (2012) shows that “organizational misfits”—people who followed career trajectories that are atypical in their organization—are more likely to communicate to workers outside their unit for sense-making. Unfortunately, our data do not allow us to distinguish between employees with different career trajectories prior to the CEO change. 12 Our communications data is at the calendar month level, so we are unable to adjust the model so that Period 0, which includes the transition month, starts on the actual transition day. In an unreported analysis, we examine the effects of the CEO transition if the CEO changes in the last 15 days of the transition month (55 firms) and find that the results of the event study are similar.

14

Third, we investigated whether the results were driven by compositional changes in the sample.

For example, the short-term drop in communications could have been generated by the exit of managers or

workers just after the arrival of a new CEO, rather than behavioral changes within incumbents. To do so,

we examine the changes in meetings and emails for employees and managers that the provider identified

as “Stayers”, i.e. whose account was active at the firm before and after the CEO turnover. In Appendix B

in Table B.2 (Stayers), we show these results for Stayers, which remain similar to the main. To support our

findings further, we summarize the changes in turnover rates by month in Appendix B in Table B.3

(Turnover).

Finally, in unreported analyses, we analyzed whether the estimated changes in communication

varied across small and large firms, for different types of reasons driving to a CEO transition (CEO death,

firing, promotion, move to a different firm or retirement), and for different types of incoming CEOs

(internal, i.e. hired from within the firm, or external CEO). We do not find strong evidence for

heterogeneous effects, though we note that our sample may be too small to estimate these differences

properly.

Communication and Performance. Finally, we investigate the relationship between communication and

cumulative abnormal stock returns, computed starting from the first month in which the CEO takes office,

and differently for firms that experienced a greater change in communication intensity in the medium term

after the CEO transition—for brevity, we refer to these firms as high communication firms.13 In practice,

we identify these high communication firms by calculating the difference in communication intensity

between Period 1, i.e. the end of the short-term transition period (months 2 and 3), and Period 3, the end of

the medium-term transition period (months 6 and 7), and generating a dummy variable taking value 1 if the

firm experienced an above-median change in communication within this time period.

13 We exclude month 0—i.e. the month in which the CEO takes office—since abnormal returns in this time period may be affected by the announcement of a CEO change rather than the CEO him/herself.

15

We present the estimation of Equation 2 in Exhibit 3. Column 1 shows that, on average, high

communication firms have different CARs immediately after the CEO change, and these differences persist

for six months after the transition. In column 2, we explore the possibility that the CARs may have been on

a different trend before the CEO change. To do so, we build a CAR measure that starts summing over

unexpected results from 6 months before the CEO change until the month in which the new CEO takes

over. The regression shows that the differences in CARs between the two groups are insignificant before

the CEO change, and we show these results visually in Figure 5.B in Exhibit 5. Next, we look at the

robustness of the relationship to the inclusion of firm fixed effects in column 3. We then repeat the same

analysis using a different CAR measure build using French-Fama factors14 in columns 4-6, finding similar

results.

All in all, these results suggest that high communication firms start to diverge in terms of

performance early on, especially in the first three months of the new CEO tenure. Yet, these differences do

not accrue over time (i.e. persistent CAR differences are entirely driven by what happens in Period 1, and

are sustained afterward). These results can be interpreted in two ways. Differences in the increase in

communication in the medium turm after a CEO turnover event may proxy for latent differences in CEO

types that become apparent to analysts and investors in the early months of a CEO tenure, and are then

factored in stock prices after this initial period. The alternative interpretation is that an unobserved, firm-

specific shock effect unrelated to the CEO type affects CARs immediately after a CEO change and that this

shock effect also induces a change in communication intensity in the medium term after the CEO takes

charge.

14 We include four factors in our French-Fama model: SMB, HML, Risk-free market returns and momentum. We share the histogram on these returns in Appendix in Figure A.4.

16

5. A Simple Model of Intra-firm Communication

In summary, our results show that a CEO change maps into significant changes in within-firm

communications. The first reaction to the CEO turnover event is a fall in internal communications, which

is followed by a medium-term increase (especially in meetings, and more specifically in meetings that

involve managers and individual contributors). We also find that differences in the medium-term increases

in communication map into different performance in the year following the CEO change.

What theories best explain the patterns we observe? One possible framework is the model of intra-

firm communication developed in organizational economics (Alonso et al. 2008, Rantakari 2008) to

formalize influential perspectives such as Hayek (1945), Sloan (1964), and Chandler (1977). In Appendix

C, we show that a highly stylized model of intra-firm communication based on Alonso et al. (2008), adapted

to the context of a CEO change, can account for the evolution of communication patterns around a CEO

transition described in the previous section.

In this framework, different units or agents within a firm pursue different objectives. The degree of

alignment around a common vision reduces the distance between different objectives and determines how

much information agents communicate with each other in equilibrium. When a new CEO arrives, there is

uncertainty about their vision for the firm, thus leading to a reduction in alignment. We should expect less

information to be communicated in equilibrium due to increased ambiguity (Weick 1995, Weick et al.

2005), which corresponds to what we observe in the data.

However, the new CEO is not a passive agent; instead, they will develop and pursue their new

vision. This involves both understating the needs and constraints of the firm and communicating new

objectives. For instance, CEOs may take into account that divisional managers have private information

(i.e., asymmetric awareness) about local conditions and the fit of these more coordinated decisions with the

external environment. Others highlight similar issues that arise with increased coordination, such as

employees engaging in non-productive or “wasteful” activities to influence decision-making (Milgrom &

Roberts 1990, Milgrom 1988, Powell 2015) which could impact communications patterns. We should

17

therefore expect a period of two-way vertical communication between the top of the organization and all

its workers. As the new vision is established, communication intensity returns to, and possibly tops, its pre-

transition level. This would explain the medium-term increase in meetings and email and why it is driven

by vertical communication.

Finally, not all new CEOs are equal when it comes to alignment (Kotter 1995). Some are better

than others at establishing and communicating a convincing vision for the firm. Those will generate higher

alignment, a larger and faster increase in communication, and ultimately higher firm performance. Our

evidence provides some reduced-form for that effect too.

While the empirical findings are consistent with the story above, it goes without saying that we

claim no causation and that our empirics can help explain other similar models of firm communication (e.g.,

Van den Steen 2017).

6. Conclusion

We use email and meeting metadata on 102 firms to study how internal communication patterns evolve

after a meaningful organizational change—a CEO transition. Then we use a theoretical model of intra-firm

communication to interpret those results. To our knowledge, this is the first time that longitudinal measures

of intra-firm communication could be analyzed and linked to meaningful organizational events across

multiple firms and longitudinally.

Using an event study research design centered on these CEO transitions, we find that CEO changes

are associated with large communication changes within firms. During the first three months after the CEO

transition, email and meetings intensity drop significantly. Around four months after the CEO transition,

there is a large increase in email and meeting communications, which more than compensates for the initial

drop. The increase in communication is mostly driven by vertical (i.e. manager to employees) relative to

horizontal (i.e. peer to peer) communication flows. These results are robust to a number of robustness

18

checks. We also find that firms that experienced a greater increase in communication in the medium term

after the CEO change also generate higher cumulative abnormal returns in their stocks immediately after

the CEO transition, and that these differences persist for six months after the event.

We see this paper as a first step in using email and meetings metadata to study the impact of

significant organizational events, such as CEO change, on internal communications patterns. Much more

could be done with data that are even richer than ours. For example, these data could be used to understand

the structure of internal communication networks with greater precision rather than using broader

aggregated measures. Furthermore, text from these communications could be analyzed to identify

discussion topics and clean the data further. Despite these limitations, we see promise in using email and

meetings metadata—typically passively collected by firms—to study unobserved aspects of the inner

workings of organizations, and more specifically, the effect that CEOs can have on internal communication

flows. We hope that our study can provide a possible blueprint to advance the exploration of these data,

while at the same preserving the confidentiality of employees and firms.

19

References

Alonso, Ricardo, Wouter Dessein, and Niko Matouschek. “When does coordination require centralization?.” American Economic Review 98, no. 1 (2008): 145-79.

Aral, Sinan, Erik Brynjolfsson, and Marshall Van Alstyne. "Information, technology, and information worker productivity." Information Systems Research 23, no. 3-part-2 (2012): 849-867.

Arrow, Kenneth J. The Limits of Organization, Norton, London and New York, 1974.

Bandiera, Oriana, Andrea Prat, Stephen Hansen, and Raffaella Sadun. “CEO Behavior and Firm Performance.” Journal of Political Economy 128, no. 4 (2020): 1325-1369.

Bandiera, Oriana, Luigi Guiso, Andrea Prat, and Raffaella Sadun. "Matching firms, managers, and incentives." Journal of Labor Economics 33, no. 3 (2015): 623-681.

Bertrand, Marianne, and Antoinette Schoar. “Managing with Style: The effect of managers on firm policies.” The Quarterly Journal of Economics 118, no. 4 (2003): 1169-1208.

Bloom, Nicholas, Christos Genakos, Raffaella Sadun, and John Van Reenen. “Management practices across firms and countries.” Academy of Management Perspectives 26, no. 1 (2012): 12-33.

Bloom, Nicholas, Raffaella Sadun, and John Van Reenen. “Management as a Technology?.” NBER Working Paper (2016) no. 22327.

Calvó‐Armengol, Antoni, Joan De Martí, and Andrea Prat. “Communication and influence.” Theoretical Economics 10, no. 2 (2015): 649-690.

Chandler, Alfred D. The Visible Hand. Harvard University Press, 1993.

Coase, R. The Theory of the Firm. Economica, 4(16), 386-405. 1937

Crawford, Vincent P., and Joel Sobel. “Strategic information transmission.” Econometrica: Journal of the Econometric Society (1982): 1431-1451.

Cremer, Jacques, Luis Garicano, and Andrea Prat. “Language and the Theory of the Firm.” The Quarterly Journal of Economics 122, no. 1 (2007): 373-407.

DeFilippis, Evan, Stephen Michael Impink, Madison Singell, Jeffrey T. Polzer, and Raffaella Sadun. “Collaborating during coronavirus: The impact of COVID-19 on the nature of work.” NBER Working Paper w27612 (2020).

Dessein, Wouter, Andrea Galeotti, and Tano Santos. “Rational inattention and organizational focus.” American Economic Review 106, no. 6 (2016): 1522-36.

Dessein, Wouter, and Tano Santos. “Adaptive organizations.” Journal of Political Economy 114, no. 5 (2006): 956-995.

Fama, Eugene F., and Kenneth R. French. “A five-factor asset pricing model.” Journal of Financial Economics 116, no. 1 (2015): 1-22.

Fama, Eugene F., and Kenneth R. French. “Common risk factors in the returns on stocks and bonds.” Journal of Finance (1993).

Gibbons, Robert, and Rebecca Henderson. “Relational contracts and organizational capabilities.” Organization Science 23, no. 5 (2012): 1350-1364.

20

Gibbons, Robert, Marco LiCalzi, and Massimo Warglien. “What Situation Is This? Shared Frames and Collective Performance.” Working Paper (2020).

Hart, Oliver, and Bengt Holmstrom. “Vision and firm scope.” Harvard University Working Paper (2002).

Hayek, Friedrich August. "The use of knowledge in society." American Economic Review 35, no. 4 (1945): 519-530.

Hoffman, Mitchell, and Steven Tadelis. "People management skills, employee attrition, and manager rewards: An empirical analysis." Journal of Political Economy 129, no. 1 (2021)

Impink, Stephen Michael, Andrea Prat, and Raffaella Sadun. “Measuring Collaboration in Modern Organizations.” AEA Papers and Proceedings, vol. 110, (2020) pp. 181-86..

Jacobs, Abigail, and Duncan J. Watts. “A Large-Scale Comparative Study of Informal Social Networks in Firms.” Management Science (2021)

Kleinbaum, Adam M., and Toby E. Stuart. “Inside the black box of the corporate staff: Social networks and the implementation of corporate strategy.” Strategic Management Journal 35, no. 1 (2014): 24-47.

Kleinbaum, Adam M., Toby E. Stuart, and Michael L. Tushman. “Discretion within constraint: Homophily and structure in a formal organization.” Organization Science 24, no. 5 (2013): 1316-1336.

Kleinbaum, Adam M., Toby Stuart, and Michael Tushman. “Communication (and coordination?) in a modern, complex organization.” Boston, MA: Harvard Business School Working Paper (2008).

Kotter, John P. “Leading change: Why transformation efforts fail.” (1995): 59-67.

Leonard‐Barton, Dorothy. “Core capabilities and core rigidities: A paradox in managing new product development.” Strategic Management Journal 13, no. S1 (1992): 111-125.

Marshak, J., and R. Radner. “Economic Theory of Teams.” Yale University, 1972.

Maurer, Indre, and Mark Ebers. “Dynamics of social capital and their performance implications: Lessons from biotechnology start-ups.” Administrative Science Quarterly 51, no. 2 (2006): 262-292.

Menon, Anoop R., and Dennis A. Yao. “Elevating repositioning costs: Strategy dynamics and competitive interactions.” Strategic Management Journal 38, no. 10 (2017): 1953-1963.

Milgrom, Paul R. “Employment contracts, influence activities, and efficient organization design.” Journal of Political Economy 96, no. 1 (1988): 42-60.

Milgrom, Paul, and John Roberts. “Bargaining costs, influence costs, and the organization of economic activity.” Perspectives on Positive Political Economy 57 (1990): 60.

Polzer, Jeffrey T., and Evan DeFilippis. “The Consequences of Collaboration Overload.” Academy of Management Proceedings, vol. (2020), no. 1, p. 21045.

Polzer, Jeffrey T., Evan DeFilippis, and Kristina Tobio. “Countries, Culture, and Collaboration.” Academy of Management Proceedings, no. 1, (2018) p. 17645.

Powell, Michael. “An influence-cost model of organizational practices and firm boundaries.” The Journal of Law, Economics, and Organization 31, no. suppl_1 (2015): i104-i142.

Rantakari, Heikki. “Governing adaptation.” The Review of Economic Studies 75, no. 4 (2008): 1257-1285.

Rotemberg, Julio J., and Garth Saloner. “Visionaries, Managers, and Strategic Direction.” The RAND Journal of Economics, vol. 31, no. 4, (2000), pp. 693–716.

21

Sah, Raaj K., and Joseph E. Stiglitz. “The quality of managers in centralized versus decentralized organizations.” The Quarterly Journal of Economics 106, no. 1 (1991): 289-295.

Schein, Edgar H. “Management of change: the case of information technology.” Information technology and the corporation of the 1990s: Research Studies 325 (1994).

Simon Herbert A. Administrative Behavior. New York, 1947.

Sloan, Alfred Pritchard. My Years with General Motors. Currency, 1964.

Srivastava, Sameer B. “Intraorganizational network dynamics in times of ambiguity.” Organization Science 26, no. 5 (2015): 1365-1380.

Steen, Eric Van den. "A Formal Theory of Strategy." Management Science 63, no. 8 (2017): 2616-2636.

Syverson, Chad. “What Determines Productivity?.” Journal of Economic Literature 49, no. 2 (2011): 326-65.

Van den Steen, Eric J. "Organizational Beliefs and Managerial Vision." Journal of Law, Economics & Organization 21, no. 1 (April 2005): 256–283

Weick, Karl E. Sensemaking in Organizations. Vol. 3. Sage, 1995.

Weick, Karl E., Kathleen M. Sutcliffe, and David Obstfeld. “Organizing and the process of sensemaking.” Organization Science 16, no. 4 (2005): 409-421.

Williamson, Oliver E. “Transaction-cost economics: the governance of contractual relations.” The Journal of Law and Economics 22, no. 2 (1979): 233-261.

Zhang, Evelyn, Brandy Aven, and Adam M. Kleinbaum. “Left but Not Forgotten: Gender Differences in Networks and Performance Following Mobility.” SSRN Working Paper (2020).

22

Exhibits

Exhibit 1 - Firm-level Event Study Regressions of Meeting and Email Measures

(1) (2) (3) (4) (5) MEETINGS EMAILS

Dependent Variable is log of:

Meetings

(count)

Avg. Duration (minutes)

Avg. Participants

(people) Emails (count)

Avg. Recipients (people)

Period -3 0.049 0.220 0.187 0.034 -0.057 (0.071) (0.077) (0.099) (0.033) (0.026) Period -2 -0.035 0.026 0.033 -0.108 -0.047 (0.071) (0.041) (0.118) (0.051) (0.020) Period -1 (Base Period) Period 0 -0.329 0.128 0.122 -0.164 -0.017 (0.105) (0.068) (0.085) (0.070) (0.031) Period 1 -0.286 0.117 0.057 -0.126 -0.016 (0.090) (0.076) (0.091) (0.055) (0.031) Period 2 0.236 0.051 0.307 0.069 0.003 (0.074) (0.056) (0.090) (0.052) (0.030) Period 3 0.227 0.175 0.513 0.111 0.025 (0.075) (0.077) (0.097) (0.047) (0.032) Period 4 0.137 0.289 0.536 0.095 0.033 (0.075) (0.086) (0.104) (0.049) (0.034) Period 5 0.268 0.269 0.379 0.108 0.029 (0.076) (0.078) (0.110) (0.052) (0.038) Period 6 -0.088 0.137 -0.081 0.014 0.035 (0.091) (0.067) (0.121) (0.083) (0.041) Observations 725 725 725 725 725 Firms 102 102 102 102 102

Notes: All columns are estimated by OLS. Standard errors are clustered at the firm level, in parentheses under the coefficient. Firm-level fixed effects are included in all models. The transition occurs during the first month of bi-monthly Period 0. Period -1, the bi-monthly period before the CEO transition, is the base period for the regressions.

23

Exhibit 2 - Firm-level Event Study Regressions of Meeting Interactions

(1) (2) (3) (4) (5) (6) MEETINGS Total Horizontal Vertical Ratio

Dependent Variable is log of:

Meetings (count)

Man-Man (count/user)

IC-IC (count/user)

All Horizontal

Vertical (count/user)

(Vertical/ Horizontal)

Period -3 0.062 0.071 0.065 0.084 -0.002 -0.030

(0.079) (0.057) (0.065) (0.070) (0.058) (0.015)

Period -2 -0.056 0.062 0.054 0.070 -0.028 -0.031

(0.077) (0.050) (0.044) (0.053) (0.055) (0.018)

Period -1 (Base Period)

Period 0 -0.183 -0.053 -0.035 -0.054 -0.092 -0.016

(0.094) (0.036) (0.044) (0.046) (0.054) (0.021)

Period 1 -0.056 -0.008 0.038 0.021 -0.005 -0.005

(0.096) (0.046) (0.045) (0.052) (0.055) (0.023)

Period 2 0.418 0.025 0.070 0.065 0.273 0.115

(0.087) (0.047) (0.045) (0.053) (0.049) (0.020)

Period 3 0.576 -0.035 0.040 0.018 0.383 0.181

(0.104) (0.056) (0.052) (0.062) (0.053) (0.023)

Observations 352 352 352 352 352 352

Firms 88 88 88 88 88 88

Notes: All columns are estimated by OLS. Standard errors are clustered at the firm level, in parentheses under the coefficient. Firm-level fixed effects are included in all models. The transition occurs during the first month of bi-monthly Period 0. Period -1, before the CEO transition, is the base period for the regressions. Regression (1) is the same measures as reported in Exhibit 1(1) but for the samples of matched observations/firms available in the meeting interactions data (352 firm-periods, 88 firms). Horizontal (1) is the mean of IC/IC and Man/Man meeting interactions. Vertical (2) is the IC/Manager meeting interactions. The ratio of vertical to horizontal meeting interactions (3) is vertical count divided by horizontal count at the firm-period level. The data underlying this table were limited to seven periods (Period -3 to Period 3).

24

Notes: All columns are estimated by OLS. Standard errors are clustered at the firm level, in parentheses under the coefficient. Firm-level fixed effects are included in models 3 and 6. The transition occurs during the first month of bi-monthly Period 0, which is used as the base period for these regressions. Firms are divided into Low and High Communications, each including 26 firms, based on whether the firm was above or below the median change in communication between month 2 and month 6 after the CEO change.

Exhibit 3 - Cumulative Abnormal Returns in High vs. Low Communication Firms

(1) (2) (3) (4) (5) (6) Dependent Variable is log of: Cumulative Abnormal Return Market-Adjusted French-Fama 3F Time Period: Post All All Post All All

Period -3 0.014 0.014 0.010 0.010 (0.031) (0.031) (0.031) (0.031) Period -2 -0.071 -0.071 -0.059 -0.059 (0.047) (0.047) (0.047) (0.047) Period -1 -0.149 -0.149 -0.126 -0.126 (0.076) (0.076) (0.076) (0.076) Period 0 (Base Period) Period 1 -0.095 -0.095 -0.095 -0.077 -0.077 -0.077 (0.046) (0.046) (0.046) (0.046) (0.046) (0.046) Period 2 -0.130 -0.130 -0.130 -0.090 -0.090 -0.090

(0.056) (0.056) (0.056) (0.057) (0.057) (0.057) Period 3 -0.139 -0.139 -0.139 -0.086 -0.086 -0.086 (0.059) (0.059) (0.059) (0.059) (0.059) (0.059) Low/High Communication (quintile) 0.033 0.033 0.028 0.028 (0.034) (0.034) (0.034) (0.034) Period -3 x High Communication -0.065 -0.065 -0.067 -0.067 (0.049) (0.049) (0.049) (0.049) Period -2 x High Communication 0.001 0.001 -0.000 -0.000 (0.070) (0.069) (0.070) (0.070) Period -1 x High Communication 0.058 0.058 0.063 0.063 (0.104) (0.104) (0.104) (0.104) Period 0 (Base) x High Communication Period 1 x High Communication 0.113 0.113 0.113 0.112 0.112 0.112 (0.050) (0.050) (0.050) (0.050) (0.050) (0.050) Period 2 x High Communication 0.132 0.132 0.134 0.136 0.136 0.139

(0.064) (0.064) (0.064) (0.064) (0.064) (0.064) Period 3 x High Communication 0.145 0.145 0.147 0.152 0.152 0.154 (0.066) (0.066) (0.066) (0.066) (0.066) (0.066) Firm-Level Fixed Effects No No Yes No No Yes Observations 290 542 542 290 542 542 Firms 42 42 42 42 42 42

25

Exhibit 4 – Event Study Figures

Notes: We plot the OLS regression coefficients from the main event regression specification on the top two graphs, reporting in Exhibit 1, across time for total meeting count (left) and email count (right) in the full sample of 102 firms. We plot the OLS regression coefficients from the meeting interaction specification across time for horizontal communication (middle left), vertical communication (middle right) on these graphs. Additionally, we look at the coefficients for the regression on the mix of vertical to horizontal meeting count (bottom left). These results are reported in Exhibit 2, and the sample consists of 88 firms. The CEO change occurs in Period 0. We use the period before the transition, Period -1, as the event study’s base period. These results include firm-level fixed effects and standard errors clustered at the firm level.

-.5-.3

-.1.1

.3.5

-3 -2 -1 0 1 2 3 4 5 6Bi-monthly Period

All Meetings

-.5-.3

-.1.1

.3.5

-3 -2 -1 0 1 2 3 4 5 6Bi-monthly Period

All Emails

-10

12

3

-3 -2 -1 0 1 2 3 4 5 6Bi-monthly Period

Horizontal Communication

-10

12

3

-3 -2 -1 0 1 2 3 4 5 6Bi-monthly Period

Vertical Communications

-.3-.1

.1.3

-3 -2 -1 0 1 2 3 4 5 6Bi-monthly Period

Ratio: Vertical/Horizontal

26

Exhibit 5 – Performance

Figure 5.A – Histogram of Communication Changes by Short and Medium Terms

Notes: This figure shows the variance of the changes in communications between month 0 and 2, and months 2 and 6 after the CEO change. Months 0-2 are considered short term, and Months 2-6 are considered medium term.

Figure 5.B – Difference between CARs of High and Low Communications Firms

(Based on medium term change in emails)

Notes: Standard errors are clustered at the firm level. Firms are divided into Low and High Communications, each including 26 firms, based on whether the change in communications between month 2 and month 6 after the CEO change was above or below the sample median.

0.5

11.

52

-1 0 1 2x

Change 0-2 months after T/over Change 3-6 after T/over

-.5-.3

-.1.1

.3.5

-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12Month

27

Appendices

Appendix A – Data

Note A.1 – Sample Construction

The sample construction involved several steps. First, the email provider gave us a sense of the time window

over which it had comprehensive email and meeting metadata (due to certain policy and data collection

system changes, the email provider had various gaps it is historical communications metadata records, and

deleted certain data on a rolling basis in certain areas to comply with regulations). We then searched for all

firms that experienced a CEO change in that period in three data sources: Execucomp, BoardEx, and Orbis.

This search led to a sample of 338 firms that experienced a CEO transition consisting of 155 publicly traded

and 183 private firms. Third, we collected auxiliary data on all these firms using public information. These

firms have on average 9,000 employees and are located in 29 countries. The majority of the firms, however,

are located in the United States (130 firms, 38%), United Kingdom (45 firms, 14%), and Canada (41 firms,

13%). We then provided this list of firms and associated firm-level data to the email provider to match our

data with their communications data.

This matching process led to a sample of 102 firms from the broader 338 firms with communication

data before and after a CEO turnover event. The matched sample includes firms located in 22 different

countries, though 39% of the firms are located in the United States. In the matched sample, 42 firms are

publicly traded. The average firm has 6,545 (SD 4,217) users, with the largest firm having almost 15,000

users. We do not have the average number of employees as measures such as employee count and revenue

levels were matched as employee count and revenue quintiles (1-5) to preserve the firm’s anonymity. We

were also not provided information on organizations of smaller than five individuals or government entities.

28

Note A.2 – Data Aggregation

The email provider sent us data at the following level of granularity across time:

• ID

• Management hierarchy (IC, M, M+)

• Tenure (Exitor, Stayer, Entrant)

• HQ (CHQ or sub)

We refer to the ID x Hierarchy x Tenure x HQ x monthnum as the cell-level. There is a user count

measure for both emails and meetings at the cell level.

For the main analysis, we sum meeting and email users at the month x id x hierarchy level (the lowest

level of granularity as we do not use Tenure or HQ data cuts. We then create a weighting variable for user

and meetings: users at the cell level divided by users at the month x id x heirarchy level. We then collapse

using the weighting variable.

Note A.3 – Performance Measures

We build two measures of performance:

• The market-adjusted CAR is calculated as (return – risk-free market rate).

• The cumulative Fama-French abnormal stock return is calculated using monthly French-Fama data

downloaded from the following website: http://mba.tuck.dartmouth.edu /pages/faculty/ken.french/.

For the Fama-French stock return measure, we use the four-factor model to estimate the residuals. These

four factors are:

• High minus Low

• Small minus Big

• Risk-free Market Returns

• Momentum

29

Since we run this model on a relatively small number of firms, we include 49 industries of firms in

training the model (also from French’s website) and then use the model to estimate abnormal returns for

the public firms in our sample.

To capture the change of performance in the mid-term, we build the pre-turnover CAR measure by

summing AR for every month before the CEO change (so the CAR for month -6 is the AR between month

-7 and -6, for month -5 it is the cumulative sum of the CAR measured at -6, and the AR between month -6

and -5, and so forth. The CAR for the post CEO change time period starts summing AR between month 0

(i.e. the month of the CEO change) and month 1 after the CEO change and so forth. Lastly, we exclude the

CAR at month 0 from our analysis since it reflects the effects of both prior and current CEOs.

Figure A.4 – CAR Histograms

Notes: These are the histograms of the cumulative average French Fama 3F one-year returns (Market(rf), SML, HML) and the OLS Market one-year return (Market(rf) for the sample of firms included in the performance regressions.

01

23

45

Den

sity

-1 -.5 0 .5FF Returns

01

23

45

Den

sity

-1 -.5 0 .5Market Returns (OLS)

30

Table A.5 – Meeting Interaction Data Availability

External IC Manager No Hierarchy

External Some Some External External

IC Some Horizontal Vertical External

Manager Some Vertical Horizontal External

No Hierarchy

Data N/A

Notes: Data summarized based on data shared by the email provider. We do not have information on the following types of interactions by employees at the firm, as we only requested proxies for horizontal and vertical, intra-firm meeting interactions between employees with recorded hierarchy information. Email information was not provided.

31

Tables A.6-A.10 – Data Summaries

Notes: We collected reason, industry, and country from all firms in the broader sample from Execucomp, Boardex, and Orbis. This data was then provided to the email provided to be paired to the communications data.

Table A.7 - Communications Data Summary

Total

Mean Med. SD Min Max

Meetings Count (avg/month) 39 29 32 1 377 Duration (minutes/meeting) 116 79 327 5 24344 Attendees (avg invited/meeting) 42 23 62 2 1470

Emails Count (avg/month) 247 223 199 1 46927 Recipients (avg employees/email) 3 3 4 1 491

Notes: Means are user weighted by average monthly email and meeting users provided by the email provider at the firm-month level for the 102 firms in the main effect analysis. Meeting count, duration, and attendees are based on calendar invites that include at least one other individual.

Table A.6 - Firm Summary

CEO Change Reason Country Industry Firms % Firms % Firms %

Within-firm Transfer 31 30% United States 40 39% Services/Fin. (SIC 60-89) 30 29% Fired/Underperformed 20 20% European Union 27 26% Manufacturing (SIC 20-39) 19 19% Other 21 21% United Kingdom 16 16% Trade (SIC 50-59) 9 9% Retirement 17 17% Canada 7 7% Other Industry 44 43% Poached/New Venture 8 8% Rest of World 12 12% M&A 5 5% Death 1 1% Total Firms 102 100% 102 100% 102 100%

32

Tab

le A

.9 -

Mee

ting

Inte

ract

ions

Sum

mar

y (p

er u

ser)

Tot

al

M

an-M

an

IC

-IC

Hor

izon

tal

V

ertic

al

Mea

n SD

Mea

n SD

Mea

n SD

Mea

n SD

Mea

n SD

Mee

tings

Cou

nt (a

vg/m

onth

)

9 6

2

2

3 3

5

4

2 2

Dur

atio

n (m

inut

es/m

eetin

g)

57

14

59

14

58

15

Not

es: M

eetin

g co

unts

and

dur

atio

n pr

ovid

ed a

t the

firm

leve

l for

88

firm

s (3

53 o

bser

vatio

ns) o

ver s

even

bi-m

onth

ly

perio

ds (c

olla

psed

from

17

mon

ths

of d

ata)

. For

thes

e ob

serv

atio

ns, w

e ha

ve m

atch

ed fi

rm-p

erio

d ob

serv

atio

ns in

the

mai

n an

alys

is. T

hese

are

dis

tinct

mee

tings

and

do

not i

nclu

de a

ny in

form

atio

n on

the

num

ber o

f atte

ndee

s. Th

e m

eans

ca

lcul

ated

abo

ve a

re b

ased

on

the

aver

age

num

ber

of m

eetin

g us

ers

per

firm

. Ver

tical

mee

tings

incl

ude

an I

C a

nd a

M

anag

er (a

nd n

o ex

tern

al o

r “no

hie

rarc

hy”

empl

oyee

s). H

oriz

onta

l mee

tings

are

the

sum

of t

wo

grou

ps: m

anag

er o

nly

mee

tings

and

IC o

nly

mee

tings

. The

Em

ail P

rovi

der d

id n

ot sh

are

sepa

rate

det

ails

on

man

ager

type

(i.e

. oth

er m

anag

ers,

seni

or m

anag

ers)

in th

is d

atas

et; t

hey

are

com

bine

d. D

urat

ion

of m

eetin

gs p

rovi

ded

only

for

IC/M

an, M

an/M

an, a

nd

IC/IC

inte

ract

ions

.

Tab

le A

.8 –

Com

mun

icat

ions

by

Man

agem

ent H

iera

rchy

Sum

mar

y

Indi

vidu

al

Con

trib

utor

s

Oth

er M

anag

ers

Se

nior

Man

ager

s

T-T

est

(p-v

alue

)

Mea

n M

ed.

SD

M

ean

Med

. SD

Mea

n M

ed.

SD

IC

/ Ot.

Man

O

t. M

an/

Sr. M

an

Mee

tings

Cou

nt (a

vg/m

onth

)

20

16

15

42

36

28

59

52

37

0.

00

0.00

D

urat

ion

(min

utes

/mee

ting)

151

85

513

10

1 77

14

5

90

77

76

0.

00

0.00

A

ttend

ees (

avg

invi

ted/

mee

ting)

36

23

50

45

24

67

46

25

68

0.

06

0.96

E

mai

ls

C

ount

(avg

/mon

th)

16

3 13

2 23

6

266

243

130

33

3 31

0 16

4

0.00

0.

00

Rec

ipie

nts (

avg

empl

oyee

s/em

ail)

4

3 6

3

3 2

3

3 1

0.

00

0.00

N

otes

: Man

agem

ent h

iera

rchy

is b

ased

on

the f

irm's

self-

repo

rted

empl

oyee

repo

rting

stru

ctur

e: IC

s hav

e no

dire

ct re

ports

, Oth

er M

anag

ers

only

man

age

ICs,

and

Seni

or M

anag

ers m

anag

e at

leas

t one

man

ager

. Gro

ups n

ot in

clud

ed in

the

firm

's fo

rmal

repo

rting

hie

rarc

hy, t

agge

d as

"no

hie

rarc

hy"

by th

e Em

ail P

rovi

der,

are

drop

ped.

Sam

ple

is th

e 10

2 fir

ms

of th

e m

ain

effe

ct a

naly

sis.

Two-

side

d t-t

est w

ith e

qual

va

rianc

es: t

he n

ull h

ypot

hesi

s is

that

the

mea

ns a

re e

qual

. IC

/Ot.

Man

: 615

mat

ched

obs

erva

tions

by

id, p

erio

d. O

t. M

an/S

r. M

an 5

64

mat

ched

obs

erva

tions

by

id, p

erio

d.

33

Table A.10 - Stock Returns Measure Summary

Cumulative Abnormal Returns Pre-transition Post-transition Mean SD Mean SD Cumulative Return (Risk Free) -0.11 0.25 -0.11 0.45 Market-Adjusted CAR -0.10 0.26 -0.11 0.27 Market (OLS) CAR -0.11 0.26 -0.11 0.27 French-Fama CAR -0.09 0.25 -0.07 0.27

Notes: Stock market return data are sourced from CSRP for the 42 firms in our sample with available data. Information on the market, risk-free (10 yr T-bill), HML, and SML rates are from Ken French’s website at dartmouth.edu. All data is daily. Market-adjusted return is the stock return minus the market return. The Market OLS model used the predicted residuals of an OLS model, including the market rate, with no constant, to predict the abnormal returns. The French-Fama model used the predicted residuals of an OLS model, including the market rate, HML, and SML, with no constant, to predict the abnormal returns. Returns are calculated at the month level and summed pre (month -6 to month 0, the transition month) and post (month 1 to month 6) CEO change.

34

Appendix B – Additional Results

Note B.1 – Monthly Data

Table B.1 shows the results of using a monthly instead of bi-monthly aggregation in the specification.

Instead of the transition occurring in the bi-monthly Period 0 (in the main results), it occurs in month zero.

The base period is the single month before the transition (Month -1), instead of the bi-monthly period,

Period -1 (in the main results). The analysis provides support that our results are not dependent on aspects

of aggregation in our specification.

Note B.2 – Stayers

To provide further support that our results were not driven by the entrance and exit of employees from the

firms around the time of the transition, we run the analysis for Stayer, defined as employees that were at

the firm before and after the transition. The email provider based this category on the date when the user’s

mailbox was created. Table B.2 shows a similar specification to the main result replicated for the subset of

Stayers in each firm.

Note B.3 – Turnover

We find that turnover remains stable in our sample. First, we show a summary of the change in the turnover

rate by month. We find no significant differences in analyzing these heterogeneous effects. The turnover

rate is calculated as: [(change in Exitors) + (change in Entrants)] / (average Stayers). Table B.3 shows the

change in the turnover from three months before the transition until 9 months after the tranistion.

35

Notes: p<0.1, p<0.05, p<0.01. All columns are estimated by OLS. Standard errors are clustered at the firm level, in parentheses under the coefficient. Firm-level fixed effects are included in all models. The transition occurs during Month 0. Month -1, the month before the CEO transition, is the base period for the regressions.

Table B.1 - Firm-level Event Study Regressions of Meeting and Email Measures (Monthly)

(1) (2) (3) (4) (5) MEETINGS EMAILS

Dependent Variable is log of:

Meetings

(count)

Avg. Duration (minutes)

Avg. Participants

(people) Emails (count)

Avg. Recipients (people)

Month -3 -0.080 0.009 -0.032 -0.093 -0.052 (0.100) (0.059) (0.162) (0.065) (0.028) Month -2 0.020 -0.023 -0.191 -0.060 -0.021 (0.094) (0.052) (0.121) (0.051) (0.028) Month -1 (Base Period) Month 0 -0.221 0.089 0.131 -0.163 -0.034 (0.107) (0.073) (0.094) (0.062) (0.029) Month 1 -0.282 0.089 -0.014 -0.202 -0.031 (0.114) (0.074) (0.107) (0.077) (0.030) Month 2 -0.366 0.016 -0.034 -0.254 -0.022 (0.105) (0.067) (0.101) (0.057) (0.031) Month 3 -0.173 0.128 -0.019 -0.124 -0.020 (0.099) (0.089) (0.106) (0.054) (0.032) Month 4 0.173 0.037 0.177 0.034 -0.009 (0.088) (0.064) (0.110) (0.046) (0.031) Month 5 0.288 0.037 0.240 0.094 0.007 (0.089) (0.066) (0.096) (0.039) (0.030) Month 6 0.300 0.056 0.444 0.108 0.017 (0.086) (0.066) (0.103) (0.038) (0.032) Month 7 0.145 0.194 0.344 0.107 0.024 (0.094) (0.094) (0.113) (0.041) (0.032) Month 8 0.110 0.175 0.410 0.085 0.032 (0.089) (0.088) (0.126) (0.043) (0.033) Month 9 0.148 0.351 0.384 0.099 0.025 (0.086) (0.093) (0.112) (0.041) (0.035) Month 10 0.083 0.208 0.113 0.096 0.026 (0.109) (0.080) (0.145) (0.044) (0.036) Month 11 0.308 0.255 0.196 0.110 0.025 (0.091) (0.088) (0.133) (0.048) (0.038) Month 12 -0.032 0.143 -0.340 0.073 0.019 (0.113) (0.074) (0.152) (0.077) (0.042) Observations 1230 1230 1230 1230 1230 Firms 102 102 102 102 102

36

Table B.2 - Firm-level Event Study Regressions of Meeting and Email Measures (Stayers Only)

(1) (2) (3) (4) (5) MEETINGS EMAILS

Dependent Variable is log of:

Meetings (count)

Avg. Duration (minutes)

Avg. Participants

(people) Emails (count)

Avg. Recipients (people)

Period -3 0.046 0.236 0.203 0.048 -0.041 (0.069) (0.074) (0.103) (0.037) (0.024) Period -2 -0.073 0.039 0.038 -0.100 -0.033 (0.075) (0.040) (0.118) (0.052) (0.017) Period -1 (Base Period) Period 0 -0.296 0.145 0.132 -0.064 -0.002 (0.106) (0.062) (0.087) (0.057) (0.026) Period 1 -0.310 0.119 0.054 -0.051 0.016 (0.093) (0.077) (0.085) (0.049) (0.027) Period 2 0.232 0.030 0.342 0.115 0.021 (0.073) (0.049) (0.088) (0.049) (0.028) Period 3 0.208 0.161 0.510 0.131 0.040 (0.072) (0.073) (0.096) (0.049) (0.030) Period 4 0.137 0.212 0.539 0.107 0.050 (0.069) (0.069) (0.104) (0.051) (0.033) Period 5 0.243 0.197 0.385 0.113 0.049 (0.071) (0.057) (0.108) (0.052) (0.037) Period 6 -0.052 0.119 -0.089 0.032 0.046 (0.086) (0.059) (0.115) (0.085) (0.044) Observations 696 696 696 696 696 Firms 96 96 96 96 96 Notes: p<0.1, p<0.05, p<0.01. This analysis is the same as the main results, however, run only on the subset of Stayers (at the firm before and after the transitions) in each firm. All columns are estimated by OLS. Standard errors are clustered at the firm level, in parentheses under the coefficient. Firm-level fixed effects are included in all models. The transition occurs during the first month of bi-monthly Period 0. Period -1, the bi-monthly period before the CEO transition, is the base period for the regressions.

37

Table B.3 - Turnover Rate Month Mean Std. Dev.

-3 -2.3% 0.093 -2 0.2% 0.007 -1 0.3% 0.012 0 5.6% 0.150 1 -2.3% 0.112 2 2.1% 0.108 3 3.0% 0.179 4 -1.6% 0.127 5 -2.0% 0.256 6 -0.7% 0.103 7 0.3% 0.098 8 -1.2% 0.099 9 1.4% 0.121

Note: The mean turnover rate is calculated as: [(change in Exitors) + (change in Entrants)] / (average Stayers).

38

Tab

le B

.4 -

Rob

ustn

ess T

ests

: Fir

m-le

vel E

vent

Stu

dy R

egre

ssio

ns

(1)

(2)

(3

) (4

)

(5)

(6)

(7

) (8

) (9

) (1

0)

Sam

ple:

Tota

l

Dep

artm

ent D

ata

B

alan

ced

Dat

a

Pre-

Ann

ounc

emen

t

Dep

ende

nt V

aria

ble

is lo

g of

:

Mee

tings

(c

ount

) E

mai

ls

(cou

nt)

M

eetin

gs

(cou

nt)

Em

ails

(c

ount

)

Mee

tings

(c

ount

) E

mai

ls

(cou

nt)

M

eetin

gs

(cou

nt)

Emai

ls

(cou

nt)

Tim

elin

e: A

nnou

cem

ent =

P0

or T

rans

ition

= P

0

Tran

s Tr

ans

Tr

ans

Tran

s

Tran

s Tr

ans

Tr

ans

Ann

Tr

ans

Ann

Peri

od -3

0.04

9 0.

034

-0

.005

-0

.016

0.04

8 0.

034

0.

037

0.

019

(0

.071

) (0

.033

)

(0.0

71)

(0.0

36)

(0

.072

) (0

.034

)

(0.0

94)

(0

.051

)

Peri

od -2

-0.0

35

-0.1

08**

-0.0

63

-0.1

22**

-0.0

35

-0.1

09**

-0.0

15

0.09

0 0.

010

0.02

8

(0

.071

) (0

.051

)

(0.0

86)

(0.0

55)

(0

.072

) (0

.051

)

(0.0

44)

(0.1

27)

(0.0

54)

(0.0

52)

Peri

od -1

(Bas

e Pe

riod

)

Peri

od 0

-0.3

29**

* -0

.164

**

-0

.169

* -0

.080

-0.3

24**

* -0

.161

**

-0

.372

**

-0.2

16

-0.1

41**

* -0

.114

(0

.105

) (0

.070

)

(0.0

91)

(0.0

51)

(0

.106

) (0

.070

)

(0.1

45)

(0.1

61)

(0.0

51)

(0.0

86)

Peri

od 1

-0.2

86**

* -0

.126

**

-0

.112

-0

.123

**

0.

096

-0.0

77

-0

.610

***

-0.1

01

-0.2

45**

* -0

.246

(0

.090

) (0

.055

)

(0.0

95)

(0.0

52)

(0

.072

) (0

.059

)

(0.1

33)

(0.1

00)

(0.0

79)

(0.1

63)

Peri

od 2

0.23

6***

0.

069

0.

365*

**

0.07

1

0.29

3***

0.

080*

0.05

9 -0

.284

* -0

.038

-0

.168

(0

.074

) (0

.052

)

(0.0

77)

(0.0

51)

(0

.073

) (0

.047

)

(0.0

63)

(0.1

44)

(0.0

34)

(0.1

00)

Peri

od 3

0.22

7***

0.

111*

*

0.29

1***

0.

092*

0.11

2 0.

107*

*

0.08

7 0.

061

0.00

1 -0

.040

(0

.075

) (0

.047

)

(0.0

84)

(0.0

51)

(0

.075

) (0

.048

)

(0.0

79)

(0.0

73)

(0.0

41)

(0.0

92)

Peri

od 4

0.13

7*

0.09

5*

0.

081

0.06

3

0.11

7 0.

095*

0.04

7 0.

132

0.01

7 -0

.014

(0

.075

) (0

.049

)

(0.0

95)

(0.0

49)

(0

.075

) (0

.051

)

(0.0

64)

(0.0

87)

(0.0

44)

(0.0

89)

Peri

od 5

0.26

8***

0.

108*

*

-0.1

08

0.05

5

0.20

7**

0.14

4***

0.12

2 0.

160*

* 0.

024

-0.0

03

(0.0

76)

(0.0

52)

(0

.126

) (0

.051

)

(0.0

78)

(0.0

52)

(0

.079

) (0

.076

) (0

.046

) (0

.088

) Pe

riod

6

-0

.088

0.

014

-0.2

14**

-0

.042

-0.1

07

0.25

9***

0.

005

0.01

7

(0

.091

) (0

.083

)

(0

.106

) (0

.111

)

(0.0

98)

(0.0

86)

(0.0

47)

(0.0

92)

Obs

erva

tions

725

725

46

9 46

9

463

463

21

3 17

3 21

3 17

3 Fi

rms

10

2 10

2

89

89

55

55

30

30

30

30

Not

es: *

p<0

.1, *

* p<

0.05

, ***

p<0

.01.

Sta

ndar

d er

rors

are

clu

ster

ed a

t the

firm

leve

l. Fi

rm-le

vel f

ixed

eff

ects

are

incl

uded

in a

ll m

odel

s. M

odel

s (1)

and

(2) a

re th

e m

ain

effe

ct, m

appi

ng to

Exh

ibit

1, m

odel

s (1

) and

(4),

resp

ectiv

ely.

Mod

els

(3) a

nd (4

) tes

t the

robu

stne

ss o

f the

mai

n ef

fect

on

the

sam

ple

of 8

8 fir

ms

that

hav

e ad

ditio

nal

data

on

com

mun

icat

ion

with

in a

nd a

cros

s fun

ctio

nal d

epar

tmen

ts w

ithin

the

firm

. Mod

els (

5) a

nd (6

) tes

t the

robu

stne

ss o

f the

mai

n ef

fect

on

the

sam

ple

of 5

5 fir

ms t

hat

have

bal

ance

d da

ta, b

efor

e an

d af

ter

the

CEO

tran

sitio

n ev

ent.

Mod

els

(7)

and

(9)

test

the

robu

stne

ss o

f th

e m

ain

effe

ct o

n th

e sa

mpl

e of

30

firm

s th

at h

ave

the

anno

unce

men

t of t

he C

EO c

hang

e m

ore

than

1 m

onth

bef

ore

the

trans

ition

occ

urs,

on th

e tra

nsiti

on ti

mel

ine

(whe

re th

e C

EO tr

ansi

tion

even

t =Pe

riod

0 an

d th

e ba

se =

Pe

riod

-1, o

ne p

erio

d be

fore

the

CEO

tran

sitio

n ev

ent),

sim

ilar t

o th

e ot

her r

egre

ssio

ns re

porte

d. M

odel

s (8)

and

(10)

use

the

anno

unce

men

t tim

elin

e fo

r thi

s sam

e sa

mpl

e of

30

firm

s (w

here

the

anno

unce

men

t of t

he p

endi

ng C

EO c

hang

e= P

erio

d 0

and

the

base

= P

erio

d -1

, one

per

iod

befo

re th

e an

noun

cem

ent.)

39

Appendix C – Simple Model of Intra-firm Communication

Subsection C.1. introduces a simplified version of this model and revisits two of its main results. Subsection

C.2. adapts the model to include the possibility of CEO transitions. Subsection C.3. derives four predictions

on communication dynamics around CEO transition by making three assumptions on the short-term and

medium-term effects of a change in leadership.

C.1. A Simple Model of Intra-Firm Communication

Definitions. A highly stylized firm is modeled as a game with three players: the Center and two Agents.

Each Agent 𝑖𝑖 has a local payoff:

πi = Ki − (di − θi)2 − δ(di − d−i)2

where 𝐾𝐾𝑖𝑖 is a constant, 𝑑𝑑𝑖𝑖 is Agent 𝑖𝑖’s decision, and 𝜃𝜃𝑖𝑖 is the local state observed by 𝑖𝑖. The 𝜃𝜃𝑖𝑖’s are normally

distributed with mean zero and variance one, and mutually independent. Besides the constant, the payoff

depends on two components:

(1) Adaptation cost, (𝑑𝑑𝑖𝑖 − 𝜃𝜃𝑖𝑖)2: how well the Agent’s decision fits the local state of the world that she

faces.

(2) Coordination cost, (𝑑𝑑𝑖𝑖 − 𝑑𝑑−𝑖𝑖)2: how well the Agent’s decision fits with the decision taken by the

other Agent.

The parameter δ measures the relative importance of coordination versus adaptation. It plays a

central role in our analysis, and we refer to this variable as the need for coordination. If the need for

coordination is high, the Agent’s payoff depends more heavily on how her decision fits the other Agent’s

decision. Each Agent cares both about her functional unit’s payoff and the other Agent’s payoff. She

maximizes:

𝜋𝜋𝑖𝑖 + 𝛼𝛼𝜋𝜋−𝑖𝑖,

40

where 𝛼𝛼 represents the degree of alignment of the Agent with the rest of the organization. When 𝛼𝛼 = 0,

alignment is minimal, and the Agent only cares about her unit’s payoff. When 𝛼𝛼 = 1, alignment is maximal,

and the Agent cares about the whole organization’s payoff 𝜋𝜋𝑖𝑖 + 𝜋𝜋−𝑖𝑖.

Communication Modes. In Alonso et al. (2008), there are two possible modes of organization and

communication. Horizontal communication occurs when the two Agents communicate with each other and

then make decisions independently. Namely, first, each Agent observes the value of her local state. Second,

Agent 1 sends a non-verifiable signal to Agent 2, and Agent 2 sends a non-verifiable signal to Agent 1.

Third, each Agent makes a local decision. Vertical communication occurs when the two Agents

communicate with the Center, who then makes decisions for both Agents (or tells them what decision to

take).15 Namely, after each state observes her local states, she sends a non-verifiable signal to the Center,

who then selects both 𝑑𝑑1 and 𝑑𝑑2.

In both communication modes, each Agent faces a tension between communicating the true value

of her signal, who will help the other player—be it the Agent or the Center—make a correct decision, and

exaggerating her own signal in order to induce the decision-maker to make a decision that is closer to the

Agent’s preference. For instance, if 𝜃𝜃𝑖𝑖 is positive, Agent 𝑖𝑖 knows that her signal is likely to be higher than

the other Agent’s signal. If signals were taken at face value, Agent 𝑖𝑖 would have an incentive to

communicate a signal higher than 𝜃𝜃𝑖𝑖 to induce the other Agent or the Center to select a higher action. This

effect is related to the one present in the celebrated cheap talk game studied by Crawford and Sobel (1982).

As in Crawford and Sobel (1982), in equilibrium, Agents communicate a partitional signal.

Although each Agent observes a continuous signal, she can only credibly communicate a coarser discrete

signal with a finite number of realizations. Each realization informs the receiver that the true signal lies in

a given interval. The residual variance after communication is a function of how coarse the partitional signal

15 There are also two hybrid configurations: (i) The Agents communicate horizontally and the Center makes (uninformed) decisions; (ii) The Agents communicate with the Center and then make decisions themselves (without getting a signal from the other Agent). However, they are obviously suboptimal.

41

is. Therefore, in equilibrium, the partitional signal’s fineness can be interpreted as a measure of the amount

of information transmitted.16

While the incentive to exaggerate the Agent’s own state is present both in horizontal and vertical

communication, the two modes of communication differ in the relative importance decision-makers assign

to local (i.e. functional unit) and global (i.e. firm-level) payoffs. As the Center always maximizes 𝜋𝜋1 + 𝜋𝜋2,

we should expect her to assign more importance to achieve coordination gains.

Equilibrium. We next present two results borrowed from Alonso et al. (2008) that will form the basis of

our predictions. The first links alignment and communication:

Proposition 1: For both modes of communication, an increase in the degree of alignment

𝛼𝛼 increases the amount of information transmitted.17

The intuition for this result is that more aligned Agents face a lower incentive to misrepresent their

information because they internalize a higher share of the cost of misleading the other Agent or the Center.

When the degree for alignment α is higher, the partitional message becomes finer and more information is

transmitted. This proposition captures the basic idea that alignment breeds trust, and we communicate more

and better when we trust each other.

This proposition applies to both modes of communication. In the vertical mode, communication occurs

between the Center and the Agents, and information transmission is higher if alignment is higher. In the

horizontal model, communication is directly between Agents and, again, it depends on alignment.

The second result links the need for coordination with the relative value of the two communication

modes:

16 A partitional equilibrium with 𝑙𝑙 intervals would require transmitting approximately 𝑙𝑙𝑙𝑙𝑙𝑙 𝑙𝑙 bits of information. 17 Proof: See Proposition 3(iii) in ADM (noting that our α is related to their λ according to α=λ/(1-λ)).

42

Proposition 2: Holding α constant, an increase in the need for coordination 𝛿𝛿 makes

vertical communication (weakly) more efficient than horizontal communication.18

This second result derives from a tradeoff identified by Hayek (1945) between centralized and decentralized

decision-making. Centralization, which builds on vertical communication, is more effective at solving

coordination problems. Decentralization, which instead exploits horizontal communication, is better at

using local knowledge, and thus at solving adaptation problems. If a firm experiences an increase in the

relative importance of coordination over adaptation, it should optimally respond by increasing the relative

importance of vertical over horizontal communication.

C.2. CEO Transition

Equipped with our intra-firm communication theory, we explore the effects of CEO transition on the

intensity and mode (i.e. horizontal versus vertical) of intra-firm communication. We distinguish between

four stages of the transition: the steady-state (when the prior CEO still runs the firm), the short-term

transition (the first weeks of the transition), medium-term transition (the first months of the transition,

excluding the very first weeks), and the new steady-state.

We make three assumptions on the evolution of our two key parameters: α and δ. First, the initial

weeks of the transition are characterized by a certain degree of confusion in the organization and a lack of

clarity over the implicit and explicit incentive structure. In the short term, alignment suffers, and each Agent

becomes more protective of her turf. Following a large organizational change, increased ambiguity creates

numerous, conflicting interpretations of the same situation (Weick 1995, Weick et al. 2005). This ambiguity

persists until the new CEO effectively communicates a singular path forward, dispelling any possible

misunderstandings regarding the firm’s mission or strategy and reducing uncertainty. Thus, we make the

following assumption.

18 Proof: See Proposition 5 of ADM as depicted in their Figure 6. For any given value of the alignment parameter (our α, their λ – see previous footnote), decentralization is always optimal for a low value of the need for coordination 𝛿𝛿. As 𝛿𝛿 increases, two cases are possible. If λ is low, decentralization remains optimal (weak increase). If λ is high, centralization instead becomes strictly better with a high enough increase in the need for coordination.

43

Assumption A: In the short term, the degree of alignment falls from the pre-transition level

𝛼𝛼0 to 𝛼𝛼𝑆𝑆𝑆𝑆, where 𝛼𝛼𝑆𝑆𝑆𝑆 < 𝛼𝛼0.

In the medium term, the new CEO takes control of the firm, selects new leaders, and sets the firm’s

strategic vision. A new leader may change the firm’s shared frame (Gibbons et al. 2020). Planned

organizational changes require increased coordination, yet inertial pressures constrain the firm’s ability to

adapt (Gargiulo & Benassi 2000, Maurer & Ebers 2006). Prior structural rigidities (Leonard-Barton 1992)

are challenging to overcome, and the repositioning costs of changing firm strategy further reduce flexibility

(Menon & Yao 2017). The communication of the strategic visions ripples through the firm when

information sharing is lower, requiring employees to update themselves and their teams on the updated

strategies. Subsets of the firm begin to coordinate their decisions based on the shared strategy, reducing

uncertainty and increasing the need to coordinate. So, we assume:

Assumption B: In the medium term, the need for coordination increases from the pre-

transition level 𝛿𝛿0 to 𝛿𝛿𝑀𝑀𝑆𝑆, where 𝛿𝛿𝑀𝑀𝑆𝑆 < 𝛿𝛿0.

Finally, in the medium term, the CEO affects alignment. The CEO has an unobservable type that

determines her ability to create the right organizational culture and implement effective incentive

mechanisms. The CEO’s type is 𝜃𝜃 ∈ {𝑏𝑏𝑏𝑏𝑑𝑑,𝑙𝑙𝑙𝑙𝑙𝑙𝑑𝑑}. The quality of management is important in multi-

divisional, decentralized firms (Sah & Stiglitz 1991), and aspects of CEO behavior and fit with the firm are

related to increased firm performance (Bandiera et al. 2015, 2020). Good managers reduce employee

turnover (Hoffman & Tadelis, 2021). Good managers can communicate their plans more effectively,

reducing greater ambiguity (Kotter 1995, Schein 1994). This reduction in ambiguity facilitates

communication. We make one last assumption:

Assumption C: In the medium term, a good CEO increases alignment relatively more than

a bad CEO: 𝛼𝛼𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔 > 𝛼𝛼𝑏𝑏𝑏𝑏𝑔𝑔.

44

C.3. Predictions

These three assumptions lead three predictions that are aligned with our results.

Prediction 1: In the short-term transition, the amount of information transmitted falls.

The first prediction is an immediate consequence of Proposition 1 and Assumption A. The reduction in the

degree of alignment leads to less intense equilibrium communication. 19

Prediction 2: In the medium-term transition, the ratio between the amount of vertical

communication to the amount of horizontal communication increases.

The second prediction derives from the assumption that the need for coordination increased after a

CEO transition together with Proposition 2. The increased need for coordination makes vertical

communication relatively more efficient. In other words, to develop a “new” organization, the CEO requires

more centralization: we thus, expect that there will be more “vertical” communication (i.e. manager to

employee and vice versa) to coordinate decision-making around the new strategy.

Prediction 3: A greater increase in communication in the medium term determines a higher

steady-state performance.

The third prediction is slightly more elaborate than the previous two, and focuses on the connection

among CEO types, increased communication, and firm performance. By Assumption C combined with

Proposition 1, a good CEO increases the degree of alignment and, hence, medium-term communication

more than a bad CEO. However, the increase in the degree of alignment also translates into a higher level

of firm performance (𝜋𝜋1 + 𝜋𝜋2 in the model). We do not know the CEO type, but we observe changes in

communication intensity and firm performance. The model predicts that the former will be correlated with

the latter.

19 One could expand our model by making communication costly. This would reinforce Prediction 1 as workers may exert less effort during the transition because they are subject to less monitoring.

CENTRE FOR ECONOMIC PERFORMANCE Recent Discussion Papers

1795 Katarzyna Bilicka Daniela Scur

Organizational capacity and profit shifting

1794 Monica Langella Alan Manning

Income and the desire to migrate

1793 Nicholas Bloom Arjun Ramani

The donut effect of Covid-19 on cities

1792 Brian Bell Nicholas Bloom Jack Blundell

This time is not so different: income dynamics during the Covid-19 recession

1791 Emanuel Ornelas Patricia Tovar

Intra-bloc tariffs and preferential margins in trade agreements

1790 Jose Maria Barrero Nicholas Bloom Steven J. Davis

Why working from home will stick

1789 Scott R. Baker Nicholas Bloom Steven J. Davis Marco Sammon

What triggers stock market jumps?

1788 Nicolas Bloom Robert S. Fletcher Ethan Yeh

The impact of Covid-19 on US firms

1787 Philippe Aghion Antonin Bergeaud Matthieu Lequien Marc J. Melitz Thomas Zuber

Opposing firm-level responses to the China shock: horizontal competition versus vertical relationships

1786 Elsa Leromain Gonzague Vannoorenberghe

Voting under threat: evidence from the 2020 French local elections

1785 Benny Kleinman Ernest Liu Stephen J. Redding

Dynamic spatial general equilibrium

1784 Antonin Bergeaud Clément Malgouyres Clément Mazet-Sonilhac Sara Signorelli

Technological change and domestic outsourcing

1783 Facundo Albornoz Irene Brambilla Emanuel Ornelas

Firm export responses to tariff hikes

1782 Gabriel M. Ahlfeldt Stephan Heblich Tobias Seidel

Micro-geographic property price and rent indices

1781 Ria Ivandić Tom Kirchmaier Neus Torres-Blas

Football, alcohol and domestic abuse

1780 Monica Langella Alan Manning

The measure of monopsony

1779 Holger Breinlich Elsa Leromain Dennis Novy Thomas Sampson

Import liberalization as export destruction? Evidence from the United States

1778 Andrew E. Clark Conchita D’Ambrosio Anthony Lepinteur

Marriage as insurance: job protection and job insecurity in France

1777 Marc J. Melitz Stephen J. Redding

Trade and innovation

The Centre for Economic Performance Publications Unit Tel: +44 (0)20 7955 7673 Email info@cep.lse.ac.uk Website: http://cep.lse.ac.uk Twitter: @CEP_LSE